提交 889176a4 编写于 作者: G Guanghua Yu 提交者: qingqing01

Fix configs in num_classes (#2905)

上级 0c91dc90
...@@ -10,6 +10,7 @@ save_dir: output ...@@ -10,6 +10,7 @@ save_dir: output
pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_vd_64x4d_pretrained.tar pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNeXt101_vd_64x4d_pretrained.tar
weights: output/faster_rcnn_x101_vd_64x4d_fpn_1x/model_final weights: output/faster_rcnn_x101_vd_64x4d_fpn_1x/model_final
metric: COCO metric: COCO
num_classes: 81
FasterRCNN: FasterRCNN:
backbone: ResNeXt backbone: ResNeXt
...@@ -76,7 +77,6 @@ BBoxAssigner: ...@@ -76,7 +77,6 @@ BBoxAssigner:
bg_thresh_lo: 0.0 bg_thresh_lo: 0.0
fg_fraction: 0.25 fg_fraction: 0.25
fg_thresh: 0.5 fg_thresh: 0.5
num_classes: 81
BBoxHead: BBoxHead:
head: TwoFCHead head: TwoFCHead
...@@ -84,7 +84,6 @@ BBoxHead: ...@@ -84,7 +84,6 @@ BBoxHead:
keep_top_k: 100 keep_top_k: 100
nms_threshold: 0.5 nms_threshold: 0.5
score_threshold: 0.05 score_threshold: 0.05
num_classes: 81
TwoFCHead: TwoFCHead:
num_chan: 1024 num_chan: 1024
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册